Aug 2024 – Jan 2025 · 5 mos · Bengaluru, Karnataka, India · On-site

• I was a part of E-statement generation team at Standard chartered. I also worked on Loan Origination process project.

Feb 2024 – Jun 2024 · 4 mos

• Developed framework to extract the relevant UPSC questions from the Database as asked by the student on Kalam chat.
• Reduced input token size per user query from 12,000 to 600 using fine-tuning techniques for LLMs and using RAG based methods
• Designed LLM based model to evaluate subjective UPSC answer scripts, providing students with relevant marks and feedback.

May 2023 – Jul 2023 · 2 mos · Chennai, Tamil Nadu, India · On-site

• Successfully refactored the UI, creating a more organized and compact table for alerts as per the client’s demand.
• Implemented new features such as selective retry and marking subscribers as ignored, enhancing the alert-sending process for
• improved user experience.
• Developed a JAVA based CSV file editor that executed transformations based on commands from the commands.txt file

Jan 2023 – May 2023 · 4 mos

• Developed a Model to detect the working status of Chimney and the Condenser from the satellite images and classify them as
• working Chimneys, Non-Working Chimneys, Working Condensing Towers and Non-Working Condensing Towers.
• Used ESR-GAN Model for generating the super resolution images and further used Faster-RCNN and YOLO V8 Model for
• detecting above classes and creating bounding boxes around them.
• Further Enhanced the accuracy by using Height Filtering and Direction Filtering (using PCA), thus removing false positives like
• building, towers, roads and trees that were missed classified as chimneys.

Sep 2021 – Nov 2021 · 2 mos

• Designed and implemented a Deep Learning Model to identify the temporal relations (before, after ,equal and vague) between the parts of a sentence.
• Used the Bert Embeddings of the complete sentence for classification to get an accuracy of 67.5 %.
• Got the set of event words along with their index and got their Bert Embeddings.
• Concatenated and added these embeddings and used them for classification to get an improved accuracy score of 71.1%.
• Implemented and trained a Siamese Network classifier on independent dataset to classify pair of words as before or after.
• Further concatenated the internal embeddings of the Event words with the hidden state from Siamese network classifier to get an improved accuracy of 76.7 %

Jun 2021 – Jul 2024 · 3 yrs 1 mo

• Enhancing Wikipedia with Biographies and autobiographies | Master’s Thesis Project|
• Proposed and developed a framework which automatically retrieves information from reliable sources like biographies and
• auto-biographies and using it, generates information which needs to be incorporated into the Wikipedia page
• Created a Chapter-Keyword Map using Key-BERT and Chapter-Year Map and merged them to form single temporal keywords
• Formed the clusters of related sentences using RAG Pipeline and Chroma as Vector Database and Mistral, Google Palm as LLMs.
• Fine-Tuned LLM Models like FALCON, LLAMA2, T5 and BART for generating the appropriate title using PEFT and QLORA Methods
• Fear Speech Detection Model
• Scraped 5M+ texts from Twitter and Gab to build a Fear/Hate speech detection model.
• Extracted word embeddings using Bag of Words, TF-IDF, and classified texts with Logistic Regression and XGBoost.
• Fine-tuned BERT for emotion vector pooling to classify texts into Fear, Hate, or Normal.
• Applied LDA Topic Modeling to identify targets of Fear and Hate speech, and created word-shift graphs with cosine similarity-based word-node connections to analyze text patterns

May 2021 – Jan 2022 · 8 mos

• Studied and verified the Blade Element Theory on the Turbine NACA 0012 aerofoil.
• Calculated the Power coefficient by taking in consideration of tip loss factor and higher values of linear induction factor.
• Wrote a python code to fully automate the process of calculating the power coefficent and induction factor for any turbine.
• Stimulated a 3 blade Horizontal Axis Wind Turbine in Ansys Fluent having Airfoils S818, S825 and S826 for the root, body and tip respectively to get the Cp value to be 0.38.
• Stimulated the wind turbine in Q-blade and got the values of Tip Speed Ratio and angle of attack at which Cp is maximum.
• Further optimised the twist value for the wind turbine to increase the power coefficient by 8 % to final value of 0.46 .

Dec 2020 – Jan 2021 · 1 mo

• Applied Data Visualization on data obtained from survey of 50,000 people from the world regarding social Relationship preferences and extracted the required data.
• Formed buckets of similar questions and designed a consistency score for each respondent.
• Eleminated inconsistent responses by making a cutoff function and reduced dimentionality of the dataset by eliminating similar responses and performed hypothesis testing on the reduced dataset